AI Researcher Calls for Tech Forecasting to Guide Policy on Frontier Progress
A post on X argues that slowing frontier AI only makes sense if governments know what extra time is for.
Entities: Joshua Achiam, Herbie Bradley
In a post on X, AI researcher Joshua Achiam argued that calls to pause frontier AI progress often skip a practical question: What should society actually do with that extra time? His answer, at least in outline, is broader technology forecasting. Achiam wrote that policymakers should map the likely “tech tree” across fields and tie national policy more closely to those projections, because AI-driven breakthroughs could, in his view, trigger social conflict, economic shocks or geopolitical instability if governments are caught flat-footed.
Combined views
5.7K
2 posts, first seen 1d ago
Reactions from ranked influencers
2 posts
Something that worries me about the general push towards a pause on AI frontier progress is that there is not a very clear thesis about what to do with extra time (beyond "solve alignment," which on the one hand I am all for, but on the other hand I think I am more optimistic than most about whether alignment, per se, is the overwhelming challenge). I have some thoughts that feel just below the surface about this - there are some things I think we *should* do with extra time if we have it, or barring "extra" time, just start doing now. I have some work cut out for me to give structure and form to these thoughts; I have only very incomplete ideas and explanations at this point. But one of the big pieces is that we need to begin forecasting the tech tree in every field in a more comprehensive way, and national policy needs to become much more closely linked to projections about where the tech tree goes from here. Substantial disruptions in the form of social conflict, economic surprises, and destabilized international relations are downstream of technology shocks - e.g. if something like a space elevator becomes possible to build on a ten year time horizon - and this, IMHO, is the sort of externality people need to prepare for as a result of advances in AI.
@jachiam0 But if it's buildable on a 10 year time horizon, isn't that long enough that we can just say "geopolitics and society will adapt over that time" so that it's not a substantial disruption? We would need there to be something buildable on a <1 year time horizon perhaps?